The data analysis was originally conducted on MicroCrunch, a DOS-based statistical package.  There are some minor descrepancies between the published results and those generated by the Stata replication commands below.  (These descrepancies seem to be a function of the difference in how the two packages handle weighted data.)  So, in Stata, the following command will, as nearly as possible, replicate the published result:

probit amicicrt petnum gainrate amprocer  lostlcrt clientrv conflict newissue natlstnd [aweight=wgt]
